Scalable-Grain Pipeline Parallelization Method for Multi ...
Learning from Your Classmates: A Multi-Method Assessment of
Transcript of Learning from Your Classmates: A Multi-Method Assessment of
Learning from Your Classmates: A Multi-Method Assessment of Classmate Peer Effects in First-Year Core Courses at
Three Colleges
Jeffrey Parker, Reed College Jon Rivenburg, Reed College
Jay Beaman, Lewis & Clark College Neal Christopherson, Whitman College
With essential contributions from:
James Grant and Benjamin Westervelt, Lewis & Clark College Jan Crouter and Ruth Russo, Whitman College
Nigel Nicholson, Reed College This paper grew out of an earlier study of classmate peer effects at Reed College, which was undertaken by Jessica Hoel with Parker and Rivenburg under the sponsorship of the Bernard Goldhammer Collaborative Grant Program. The current research is part of a project supported by an Outcomes and Assessments Grant from the Teagle Foundation. Nigel Nicholson, Ruth Russo, Ben Westervelt, Jay Beaman, and Neal Christopherson are members of the Teagle project research team and contributed, directly and indirectly, to this research. The project also benefited from outstanding research assistance over several years from numerous students at all three colleges, including Nick Chandler-Klein, Kathryn Deane, Christa Gibbs, Ranganai Gwati, Joe Hand, Ben Keefer, David Kling, Joshua Lingerfelt, Morgan Matthews, Ramon Mitchell, Daniela Morar, Dylan Smith, Michael Turner, Izarra Varela, and Liza Wickline.
1
Introduction
Our study explores classmate peer effects—the effects of classmates on a student’s learning—
in required first-year core courses at three selective Northwest liberal-arts colleges. We examine
classroom peer effects with two complementary approaches, one based on quantitative evidence and
the other based on instructor interviews. The statistical analysis looks for the effects of measurable
classmate characteristics (such as admission credentials) on measurable academic outcomes (grades
and persistence). The interviews provide a rich contextual background that illuminates the results of
the statistical analysis and suggests important channels of peer effects that we were not able to
explore quantitatively with the available data.
To preview our main findings, the instructors we interviewed were strongly convinced that
peer effects are prominent in discussion-based core courses. However, the most striking consensus
was that peer effects related to students’ attitudes and classroom behavior were more important than
those associated with their abilities. While having high academic aptitude of the kind that the SAT
attempts to measure was viewed as a positive peer characteristic by the instructors, most were clear
in emphasizing behavioral rather than ability-related peer attributes. Our quantitative analysis was
restricted to student characteristics available in the college databases, which was restricted to such
admission measures as SAT scores, high-school records, and admission-office ratings. We found no
effect of these aptitude-based classmate characteristics on student learning at any of the three
schools. This suggests that future peer-effect research might be productively directed at collecting
data on the attitudes and personalities of incoming students and examining the effects that these
characteristics have on classmate learning.
Peer Effects
Peer effects are influences of an individual’s peers on his or her behavior. They may occur in
many contexts; for example, co-workers’ productivity may influence one’s own, neighbors’
consumption decisions may cause one to consume more or less, one may be tempted to drink or
smoke more if one’s friends do. We are concerned with how peer effects among college students
affect their academic achievement.
Leaders of colleges seem to believe that peer effects are important. Colleges that are in a
position to do so admit students selectively, picking those thought to have the most academic
potential. Many colleges attempt to improve the academic quality of their student bodies by offering
merit-based financial-aid awards to students who are especially promising. Among several possible
2
explanations for the pervasive aspiration to selectivity and the frequent practice of merit aid, peer
effects relate most closely to the pedagogical mission of a college.1 To the extent that having “better
peers” gives students a better education, the goal of attracting an excellent student body aids
attainment of the goal of educating students as effectively as possible. In the presence of positive
ability-based peer effects, Rothschild and White (1995) demonstrate that offering merit aid to
excellent students can enhance the efficiency of the education system by compensating students for
the “services” that they provide in enhancing their peers’ learning.
The student interactions leading to peer effects occur on many levels: roommates, dorm-
mates, friends, lab partners, classmates, colleagues in activities, and myriad other interactions. Some
student peer effects are broad and occur at the level of an institutional as a whole. For example, an
excellent overall student body allows the faculty to offer challenging courses that elevate the level of
students’ learning. Institution-wide peer effects may even extend beyond a student’s contemporary
peers, as for example when the value of his or her degree is enhanced by the past success of the
college’s alumni.
In contrast, some student peer effects occur at the individual or class level. If a student’s
roommate has poor study habits, it may distract the student from studying and lower his or her
academic performance.2 Classmates who contribute clever insights to class discussions or who are
effective study partners may improve the learning of others in the class.
In order to distinguish statistically between the effects of peers’ characteristics and the effects
of a student’s own characteristics, it is necessary to find situations in which peers are assigned either
randomly or according to observable criteria that can be used as controls.3 Most studies of classroom
peer effects have focused on primary education, presumably because it is more feasible to undertake
controlled studies with random peer assignment in elementary schools where the both curriculum
and students’ academic backgrounds are relatively homogeneous. Among the studies examining
classroom peer effects among younger students are Henderson, Mieszkowski, and Sauvegeau (1976),
Epple and Romano (1998), Hoxby (2000), Zimmer and Toma (2000), Boozer and Cacciola (2001),
Gaviria and Raphael (2001), Vandenberghe (2002), Hanushek et al. (2003), McEwan (2003),
Rangvid (2003), Robertson and Symons (2003), Zimmer (2003), Angrist and Lang (2004), Dills 1 Among the other rationales for these phenomena, high-ability students might be easy or more fun to teach
and their subsequent success might bring prestige (and financial support) to the college. 2 For example, Stinebrickner and Stinebrickner (2008) find that students whose roommates brought a video-
game system to campus studied less and earned lower grades than those whose roommates did not, controlling
for the student’s own characteristics. 3 See Manski (1993).
3
(2005), Lefgren (2004), Ammermueller and Pischke (2006), Burke and Sass (2006), Ding and Lehrer
(2006), and Kang (2007). Although the evidence is not unanimous, many studies find support for the
hypothesis that young students learn more when they have peers of higher ability.
Higher education peer effects are difficult to measure because it is usually infeasible to do a
controlled experience with randomized classmate selection. Even “natural experiments” in which
students’ peers are selected randomly or according to observable criteria or criteria unrelated to
academic outcomes are uncommon. Most higher-education studies have examined roommates
because the process of roommate selection for first-year students often conforms to the natural
experiment conditions above. Studies based on residential peers include Sacerdote (2001),
Zimmerman (2003), Winston and Zimmerman (2004), Stinebrickner and Stinebrickner (2006),
Foster (2006), Lyle (2007), Kremer and Levy (2008), and Stinebrickner and Stinebrickner (2008).
These studies have found some evidence that more able roommates are beneficial; the most recent
studies have suggested that the mechanism of roommate peer effects may operate through study time
and alcohol use, which are more closely related to the attitudes and personalities of roommates than
to their innate academic ability.
Project Overview
As noted in the introduction, this study of classroom peer effects involved two related
methods of analysis: a qualitative study consisting of conversations with experienced core-course
instructors and a quantitative study of the statistical relationship between measurable characteristics
of classmates and measurable student outcomes. We view these two components as strong
complements.
The quantitative analysis allowed us to look for formal, statistical evidence of classmate peer
effects. However, the data that could be collected limited the student characteristics and outcomes
that could be examined statistically.
The qualitative study, though it did not allow the same degree of formality, enabled us to
move beyond the limitations of the data. The qualitative examination allowed us to peek inside the
peer-effects box to explore a range of peer-interaction phenomena, albeit through the lenses of
particular professors’ recollections. Thus, one aim of the qualitative study was to provide a richer
context through which to interpret the statistical results. Another was to identify promising aspects
of peer effects that might be approached statistically through collection of additional quantitative
data.
4
The statistical analysis is summarized briefly below, followed by a more detailed
examination of the interview part of the study. More details of the statistical analysis can be found in
Parker et al. (2010).
Focus on first-year core courses
We examine the effects of classmates in required, first-year, core classes at Lewis & Clark,
Reed, and Whitman Colleges. Several reasons motivate the choice of core classes:
• These courses are required of all entering first-year students, so we need not worry about
students with particular characteristics choosing to take (or not to take) the courses. We can
therefore claim to be studying the entire (non-transfer) student populations of the three
schools rather than an idiosyncratic subset.
• Student self-selection of classmates, whether explicit or implicit, may introduce bias into
quantitative and qualitative assessment of peer effects. When students and their peers are
similar in dimensions such as interests, motivation, and ability that cannot be perfectly
measured (and thus for which we can’t perfectly control), it is impossible to distinguish
statistically between the effects of a student’s own characteristics and the effects of the peers.
Because of this problem, most statistical analyses of peer effects in higher education have
looked at roommate rather than classmate peers. In the core courses that we study, students
are assigned into sections of roughly homogeneous size by a process that is largely or wholly
random. This allows us to avoid one of the most difficult problems in examining peer effects.
• Core classes involve teaching students the process of learning in a central way alongside the
teaching of course content; students learn the reading, writing, and discussion skills it takes
to be a college student as well as learning the core material. This claims for these courses a
central position in the curriculum, which in turn suggests such core courses have an outsized
influence on a student’s academic career.
• Classmate peer influence seems likely to be a significant factor in class discussions, which are
the norm in these core courses. While there are certainly other contexts in which such
students learn (study groups, dorm discussions, etc), class discussion figures more
prominently in such classes than in homework- and/or lecture-oriented classes. Classmate
interaction should thus be more observable in these classes, both qualitatively (by the
professor) and quantitatively (because the data can be gathered specifically on the class).
5
Looking at core courses naturally leads us to emphasize the kind of peer effects that arise in
classroom discussions. This is not to deny that there are other forms of significant peer interaction,
and we did gather some incidental evidence about non-core courses from some of our interview
subjects. However, extending the analysis beyond the core-course environment lies beyond the scope
of the present study.
Courses Studied
The study looks at three selective liberal-arts colleges located in the Pacific Northwest: Lewis
& Clark College and Reed College, both in Portland, Oregon, and Whitman College in Walla
Walla, Washington. Although there are important differences among the schools, all three are small,
selective, and strongly residential, all three offer curricula dominated by traditional liberal-arts
disciplines, and most importantly for our purposes all three require first-year students to take a
discussion-based core course with a common syllabus. Also crucial for our analysis was access to
adequate historical records of individual student characteristics and achievement, which was
facilitated by the institutional research staff at all three colleges.
The core courses we study vary in topical content and to some degree in format (although all
emphasize student discussion), but they share common goals. Central aims of each school’s core
course include:
• Teaching students to read carefully and think critically,
• Teaching students to write clearly, and
• Teaching students to learn from and contribute to an intellectual conversation.
In each case, the syllabus for the course was either identical across sections or had extensive shared
content. In some cases, exams or paper topics were common across sections. Table 1 summarizes
some characteristics of the three core courses.
Statistical Evidence on Peer Effects
Our statistical approach to examining peer effects uses a regression equation of the form
,i i i iy X P u= α + β+ (1)
where yi is an academic outcome measure for student i, Xi is a set of controls for student i’s own
ability, and Pi measures characteristics of student i’s peers. The coefficient vector β measures the
effects of the peer characteristics on the student’s outcome, controlling for the student’s own ability.
6
The most common peer-effects hypothesis is that some or all students benefit from having
more able peers. This is tested by including one or more measures of peer ability in P and examining
whether the corresponding elements of β are positive. Other peer-effect hypotheses are less common
but can be examined within the framework of equation (1). For example, by including a measure of
the dispersion of peer abilities in P one can test whether students gain or lose from having a wider
range of abilities among peers. By including a measure of the number (or share) of peers who have
ability levels similar to student i, one can test the “like-me” hypothesis that students benefit from
having peers that have a similar ability level to their own.
This study uses classmates in core courses as the peer group. Our outcome measures are
grades in courses other than the core course, which is omitted because variations in peer quality are
likely to bias the relationship between learning and core-course grades by affecting the instructor’s
curve. We control for admission credentials of the student and for some non-peer elements of the
student’s core-course experience—in particular, for core-course instructor and year. Our various
measures of peer characteristics attempt to capture relevant properties of the distribution of
classmate abilities.
Variables and data
We examine five grade-point average outcome measures to test for core-course peer effects,
all of which exclude the core course to avoid confounding peer effects with the effects of the
instructor’s grading curve.4 First we use overall cumulative GPA, which includes all courses (except
core) taken at the college.5 Next, since the effects of the core-course experience may be more
important during the year in which the core course is taken and the year after, we examine first-year
and first-and-second-year GPA.6
4 Elimination of the “curve” effect in the core course is crucial to the identification of peer effects. If instructors
grade on a curve, then students with high-quality core-course peers will tend to get lower grades in the core
course, even if they learn more through the presence of excellent peers. Excluding the core-course grade
prevents this effect from confounding our peer-effect estimation. 5 All GPA measures exclude transfer work and work taken on overseas programs not taught by regular college
faculty. Courses taken on a pass-fail basis and courses from which a student withdrew are also excluded. All
courses are evaluated on a 4.0 scale with 0.3 added (deducted) for a plus (minus) grade, so an A– counts as 3.7
and a B+ as 3.3. 6 The first year is taken to be the fall semester in which the core course is begun and the following spring
semester. The second year comprises the fall and spring semesters of the next academic year. For a student
who took a year off between freshman and sophomore year, only the freshman year would be counted in the
7
Finally, the core-course experience may have a stronger impact on courses in related subjects
or that use related teaching methods than in other courses. We use a narrow and a broad version of a
core-related GPA to measure success in these courses. The narrow measure consists of courses in
fields directly related to the disciplines in the core course.7 The broader measure includes all courses
except those in the natural sciences and mathematics.
We examined numerous potential measures of peer ability and numerous aspects of the class
distribution of these measures. Because the available univariate measures of student ability (SAT
scores, high-school GPA and class percentile, and admission-office ratings) were not highly
correlated with one another, we prefer to measure the academic aptitude of a student entering the
core course by a linear combination of these univariate measures rather than any single measure.
The linear combination we choose is based on a regression of cumulative college GPA on the set of
measures, so our academic aptitude variable is a student’s predicted GPA conditional on his or her
admission-file characteristics.
Missing data
College databases (and indeed colleges themselves) are not operated for the convenience of
subsequent analysis by economists. One manifestation of this is the incomplete information available
on many students. In some cases this occurs due to external constraints such as high schools that do
not report their students’ class ranks. Sometimes it results from institutional policies such as making
submission of admission credentials such as SAT scores optional. In other cases it happens when
students and applicants fail to answer questions about their ethnicity.
Approximately 70 percent of the students in our sample have complete data for the variables
in which we are interested. Missing-data problems are not uncommon in economics and in many
applications we can simply ignore the missing cases and estimate our regressions using the sample
that is complete, provided that the selection of available cases did not introduce bias. The missing-
data problem is more severe for us, however, because our goal is to characterize the distribution of
student abilities in an entire class. If we have data on all but one or two of the students in a class
section, our picture of the class distribution of abilities is likely to be quite accurate. But with one-
third of the average class section missing, the description of the class ability distribution based solely first and second years. We chose this definition to maintain temporal proximity of the courses whose grades
are measured to the core course. 7 We solicited the opinions of the core-course chairs at each campus in making these judgments. The courses
deemed to be closely related were largely courses in history, literature, philosophy, religion, the arts, and
related fields.
8
on complete cases would be useless. In order to obtain a predicted GPA value for all or nearly all
students in each class, we must employ one or more of several methods for estimating models with
missing cases.
For most instances of missing data, such as a high-school not reporting class rank, we can
safely assume that the data are “missing at random”—the fact that the variable is missing is not
related to the value of the variable. In this case, the method of multiple stochastic imputation is
appropriate.8 The one notable except is SAT scores at Lewis & Clark College. Because Lewis &
Clark accepts but does not require SAT scores from applicants, we think that students with low SAT
scores are more likely to be missing from the data set than students whose SAT scores are high.
With data “not missing at random,” we must estimate separate regressions for two sub-samples, one
with SAT scores and one without.
Statistical results
Before testing the effects of specific characteristics of classmate distributions, it is useful to
examine whether our outcome measures tend to be correlated among members of individual class
sections. Analysis of variance applied to our GPA outcomes consistently shows strong statistical
significance for core-course section, indicating strongly that something about the core course
experience does have a significant effect on academic performance in other courses.9
Of course, the core-course effect could be due to many causes, including the effectiveness of
the instructor. But even controlling for the identity of the instructor, year of entry, and individual
student characteristics, the core-course section is still statistically significant at two of the three
colleges.10
Clearly, something about the core-course classroom seems to be affecting our outcome
measures in a statistically significant way. The next step is to search for any characteristics of the
peer distribution that might be the source of that effect. We were unable to find any measurable
characteristics of the classroom distribution of academic credentials that had a statistically (or
practically) significant effect on our outcome measures.
Table 2 shows the estimated coefficients on classmate mean predicted GPA. The dependent
variables in these regressions are the five GPA outcomes shown in the left column. In addition to 8 See Rubin (1996) for more information about this method. Our statistical results were generated using Stata
11, which incorporates algorithms based on Royston (2005). 9 Of the fifteen tests statistics (for five outcome measures across three schools), only one has a probability value
above 0.01. For details, see Table 5 of Parker et al. (2010). 10 These results are in Table 6 of Parker et al. (2010).
9
peer mean aptitude, the regressions include controls for the student’s math and verbal SAT scores
(except Lewis & Clark no-SAT sample), high-school GPA, high-school percentile, admission rating
(not available for Whitman), and batteries of dummy variables for gender, financial-aid status,
ethnicity, foreign/domestic student, year of entry, core-course instructor, and area of major (to
control for differing grading standards across majors). The coefficients on the student admission-
credential controls (not shown in Table 2) have the expected signs and strong statistical significance.
None of the estimated coefficients on the peer mean aptitude is statistically different from
zero; many are negative and many are positive. From these estimates, it appears that any peer effects
resulting from mean classmate abilities in the core courses have no effect on students’ subsequent
grades.
Moreover, these results fairly represent the large number of variations that we estimated. We
examined the effects of the dispersion (standard deviation or interquartile range) of the peer
distribution, the representation of ability quintiles in the class distribution, and many other measures
without finding any attribute of the ability distribution that had a systematic effect on student
performance. We examined peer ability distributions based solely on SAT composite scores and on
admission-office ratings (where available); once again, there was no evidence of ability-based peer
effects. We performed regressions for sub-samples of students separated by predicted academic
ability, gender, and other characteristics without finding any significant effects of classmate peers.
For more details on these tests, see Parker et al. (2010).
Evidence from Instructor Interviews
As discussed above, the instructor interview part of our study was intended as a complement
to the statistical analysis. We hoped that the interviews would help explain the mechanisms
underlying any peer effects that appeared in the statistical analysis and suggest productive future
directions for statistical research. Given that our regressions suggest that there are no important peer
effects associated with classmate ability, we look to the interviews to explain why.
Interview methodology
We gathered information for the qualitative study through interviews with core-course
instructors at all three schools. Our aim was to conduct ten interviews at each campus. We
succeeded in conducting ten personal interviews at Reed and Whitman, plus one additional
instructor at Whitman who was unable to keep the interview appointment but answered questions in
writing. At Lewis & Clark, we were able to complete only eight of the ten scheduled interviews due
10
to illness and scheduling conflicts. Thus, we have in-depth responses from 29 core-course instructors
at the three schools, in addition to the insights of the three team members who regularly taught and
chaired these courses.
In selecting the interview subjects from the pool, we sought breadth at each school in age,
sex, and home department. Although most of our interviewees had taught the course many times,
we also wanted a few younger faculty members who were relatively new to the course to see if their
“freshness” led to any useful insights. While most interviewed faculty members were tenured, a few
were untenured assistant professors and some were non-tenure-track instructors who have
specialized for many years in teaching the core course. To achieve field breadth, we interviewed
instructors from a wide variety of academic departments.
In choosing the size and composition of the project team to conduct the interviews, we
considered several important criteria. First, it seemed essential to have at least one person from the
interviewee’s own college present to provide a local context for the questions and a familiar face to
the project. Second, we wanted to have at least one person take part in all of the interviews to assure
that a similar interview protocol was followed for all sessions and so that responses across
individuals and across colleges could easily be compared. Finally, we wanted the interview team to
be small enough to allow the interviews to take place in faculty offices.
These criteria seemed best served by an interview team of two, consisting of the principal
investigator (for consistency and comparison) and the project team member from the local core-
course faculty (for local expertise and familiarity).
To allow a mid-course refinement of the procedure and questions, we conducted the
interviews in two batches. Three faculty members at each school were interviewed during March
and early April 2007 in a set of “pilot” interviews.
Our goal was to complete ten interviews at each college. We had a pool of ten or eleven
candidates at each school. The interview procedure used in the pilot interviews worked well and
only slight changes to the questions were deemed desirable. We then proceeded with the remaining
final batch of interviews in May 2007.
We wanted the interview subjects to have the opportunity to reflect in advance of the
interview on their experience with peer interactions and to consult any records that might aid their
recollection. To facilitate their preparation, we sent a set of advance questions to each interviewee;
the questions are shown below:
• What do you think are the most important collective characteristics of the students in a class section that help students learn from each other?
• Can you think of particular class sections that worked well to promote the understanding of class members? What were their characteristics? Were these class sections equally effective
11
for all members of the class or did the effectiveness vary across student characteristics (strong/weak, male/female)?
• Can you think of 2 to 3 students over your teaching career who seemed to promote the understanding of other students? What were their characteristics?
• Were any of these students effective in promoting a positive class atmosphere but were not themselves outstanding scholars?
• What specific student behaviors have most enhanced the learning of peers? What behaviors have detracted the most from peer learning?
• If you had to name a single “model” student from the class, whom would you name? Why?
Interviews were conducted either in the interviewee’s office (at Reed and Lewis & Clark) or
in the core-course team member’s office (at Whitman). We began the interview by telling the subject
that we would ask permission before quoting their remarks and asking for permission to record the
interview, which was granted by all.
We then described our interests with the following text: “We are interested in how the
characteristics of a student’s peers affect success in learning. We wish to define success relative to each
student’s own ability or potential, so successful peer interaction allows a student to learn more than they
might be expected to otherwise. We are particularly interested in how peers’ personalities or abilities
work to promote or hinder learning success relative to their own potential. These effects could be the
result of individual actions on the part of a peer or could occur through larger group or full-class
interactions.” We concluded the introductory phase of the interview by asking the instructor to
avoid identifying individual students by name, using pseudonyms or other mechanisms to preserve
anonymity.
The body of the interview consisted of a conversation prompted by our prepared questions.
Most questions were posed by the local core-course team member with the principal investigator
recording detailed notes and following up or filling in with occasional questions.
The questions that guided our interviews are shown below with a rough topical organization:
• Introductory Questions o What criteria do you think are most important in assessing student learning in the core
course? In other words, what outcomes will a successful student have achieved through the course? Are these success criteria reflected in students’ course grades?
o Do you think that the quality of his or her peers affects a student’s learning? • Successful and Unsuccessful Class Sections
o What collective characteristics of the students in a class section do you think are the most important in helping students learn from each other?
o Can you think of particular class sections that worked well to promote the learning of class members? How would you characterize the mix of students in these classes? Why were they effective?
12
o Were there particular class sections that worked particularly poorly in promoting learning? How would you characterize their mix of students? Why were they ineffective?
o Were these class sections equally effective or ineffective for all members of the class or did the effectiveness vary across student characteristics?
• Beneficial and Detrimental Peers o Can you think of two or three students over your teaching career whose presence in the
class seemed to promote the understanding of other students? What were their characteristics?
o Were the students who were effective in promoting a positive class atmosphere also themselves outstanding scholars?
o Can you think of two or three students over your teaching career who benefited from peers more than others? What were their characteristics?
o What specific student behaviors have most enhanced the learning of peers? o What behaviors have detracted the most from peer learning?
• Goals of Students in Core Course o What do you think are the main goals students hope to achieve in the core course? Do
they vary across students? To what extent do you think classmate peer interaction is promoted by having similar educational goals?
• Peer Effects and Specific Groups of Students o Have you observed differences between male and female students in how they learn from
each other? o Does the presence of members of minorities groups, international students, transfer
students, or older students affect how the class learns? • External Effects on the Core Classroom
o Do the nature of the classroom and/or the time of day have significant effects on student learning and peer interaction?
• Mix of Abilities o In thinking about how the mix of student intellectual abilities affects learning, do you
think that a diverse mix of abilities is better for weak students or that they learn best when most of their peers are close to their ability? How about for stronger students?
• Peer Interaction Outside of Class o Because the core course is taught from a common syllabus, there are opportunities for peer
interactions among students from different sections. Have you observed such interactions? How beneficial were they for the students?
• Final question o What other things might we have asked or that you think are important in better
understanding classmate effects on student learning?
13
These questions were used by the interviewers to prompt the instructors to explore the broad
question of peer effects, not as a strict rubric for scoring responses or bounding the conversation.
During the interviews, instructors often went outside the boundaries of the specific question that was
asked, either anticipating later questions or raising issues that were not covered in our set of
questions. We encouraged this and adjusted our subsequent questioning to avoid repeating topics
that had been covered earlier.
After performing and analyzing the interviews, we invited the interview participants and
faculty representatives from other colleges and universities with common-syllabus core courses to
discuss our tentative findings at a conference, held at Reed College in May 2008. The conference
featured enthusiastic discussions among about forty participants, leading us to refine our
understanding of peer effects.
Interviews with instructors: Results and analysis
Are peers important?
Yes. The vast majority of interviewees felt that peers can have a strong positive or negative
influence on their classmates’ learning. One stated, “I think the chemistry of an individual section
has everything to do with whether or not some students go far.” A remarkable number of the
interviewed instructors admitted, however, that they had not thought much about peer effects prior
to our study.
Most of the instructors were able to come up with clear examples of individual students who
had influenced their classmates and examples of individual sections that were more and less
successful. One example was from an instructor who had taught multiple sections in the same
semester. A student from the weaker section who happened to attend the stronger class would
perform at a higher level than he or she normally did in the regular section.
Attitude vs. aptitude
Given the prominence of aptitude measures in our statistical work, we were naturally very
interested in how a student’s actual or potential academic ability, as measured by the student’s
grades or admission credentials, connected with his or her influence as a peer. Do students gain from
having more able peers who achieve higher grades and enter with more impressive credentials? Do
they gain from peers who are close to their own level of academic standing? Is there some other
complementarity between peers?
14
However, a strong consensus emerged from our interviews that characteristics of the
student’s personality and attitude had more effects on peers than did the student’s raw academic
ability as measured by grades or admission numbers. While most agreed that the most beneficial
peers had been average-or-better students, it was notable that in only a few cases were the stellar
peers also the strongest when measured on the usual grading scale. For example, the most
outstanding peers were often not excellent writers.
The importance that the instructors attached to students’ personalities may reflect the
emphasis that they placed on classroom discussions in assessing peer effects. A central premise of
class discussion is that all students can learn from the group: that the interaction of individuals with
varying backgrounds and opinions provides a richness of understanding that is not possible to
achieve individually. Thus, instructors consistently lauded peer behavior that advanced the collective
expression of ideas and denigrated behavior that impeded it.
Academic ability surely affects the content of a student’s classroom contributions: smarter
and better prepared students would be expected to have greater insight on the readings and to thus
make comments that have greater academic merit. While the contributions of these students to the
discussion may advance the class toward academically profound answers, they do not necessarily
advance the active participation of their peers in the conversation and so may not contribute to
achieving the richest and deepest collective responses to questions. Indeed, they may have quite the
opposite effect if their contributions are voiced in a superior, dismissive, belittling, or intimidating
tone.
The instructors we interviewed consistently ranked as the most effective peers the students
whose contributions to class discussions seemed to open up space for the contributions of others.
While these students were often academically very strong, we heard about many students who were
not academically outstanding but who seemed to have a strongly positive effect on their peers. A
central conclusion from our analysis of the interviews is that instructors think attitude affects the
quality of peer effects more than aptitude.
Good peer behavior
There were many peer characteristics that were mentioned by many or most of the core-
course instructors we interviewed. The most obvious (and probably universal) beneficial behavior
was simply attending and being prepared for class. In small discussion sections, absences are easily
noticed by both instructors and peers. Students who fail to participate because of inadequate
preparation also drag the class down, as do those who attempt to dominate despite being
unprepared.
15
As noted above, most of the positive peer characteristics related to students’ personalities
and their attitudes toward the course and toward academic study in general. Many of these
characteristics overlap and one would often expect to find students exhibiting one characteristic to
have others as well.
• Attendance and careful preparation for class were at or near the top of most lists. Indeed,
this positive characteristic was obvious enough that few instructors felt the need to dwell on
it.
• Maturity, leadership, and concern for the progress of the class were mentioned in various
forms by many instructors. Students who are genuinely concerned with how the class is
going and who have the personal tools to take the lead in facilitating peer interaction seem to
be priceless assets in a discussion-based classroom. Self-confident, humble, energetic,
empathetic, and cheerful were mentioned as desirable traits often seen in these students.
• General intellectual curiosity was almost universally listed as a desirable peer characteristic.
Students who like to think about abstract questions and learn for learning’s sake tend to
bring a positive influence to class discussions. They are good role models for others who are
less convinced of the merits of the intellectual enterprise. Several interviewees used the term
“playfulness” with ideas to describe beneficial peer behavior.
• Respect for the opinions of other students was another near-universally mentioned
characteristic. Students who believe that their peers’ insights are valuable usually listen to
them carefully and respond thoughtfully and in a way that invites further comment. This
kind of participation leads to improved class dynamics. One instructor noted that students
feel like they are taken seriously when one of their peers (or the instructor) subsequently
refers to or builds on their comments, even if they end up disagreeing. Such behavior
validates the student’s contribution to the discussion and encourages further participation.
• Demonstrated enthusiasm for the core course and its subject matter was often described as
important. Even students with general intellectual curiosity are not necessarily excited about
the subjects that dominate the reading lists of core courses. Science majors, in particular,
were sometimes identified as (at least initially) uninterested in the humanistic core fields.
Excitement about the subject makes good peers in at least two ways: (1) such students are
very likely to come to class well prepared and (2) they may convey their enthusiasm to
others. Some instructors perceived the pervasive local attitude toward the core course to be
quite negative; these instructors were likely to view enthusiasm for core as a positive
16
characteristic. (The degree to which instructors were concerned about the prevalent student
attitude toward the core course varied considerably across schools.)
• Openness to new ideas and approaches was mentioned by many instructors. Those with a
willingness to revise their opinions—to see through the comments of others that their initial
ideas were mistaken—stand to gain more from peer interaction and may contribute more to
it. By contrast, closed-minded or dogmatic students often find little basis for productive
discussion with their peers.
• A cooperative rather than an overly competitive attitude toward academics was
highlighted by some interviewees. Cooperative students were more interested in advancing
the class conversation toward an improved shared understanding; overly competitive
students often attempted to “score points” through their comments or to convince the class
to adopt their opinions, which can shut down further discussion. One instructor
characterized cooperative peers simply as “civil people.”
• Willingness to speak in class was important, though most instructors felt that shy students
could contribute positively without making frequent contributions. “Extroverts that are not
narcissists” was how one characterized the ideal peer personality. Several instructors noted
that students with backgrounds in acting were often classroom catalysts because they were
uninhibited participants. Our interviewees’ opinions about the quiet peer are discussed in
more detail below.
• Taking chances was viewed as important by many instructors. When recalling excellent
peers, many described situations in which a student “exposed himself” by “being puzzled
out loud” or “talking through an idea in front of the other students.” By putting such an idea
or question in front of the class, the student provided a focus for productive conversation
among his or her peers. One instructor pointed out that by being uncertain or even wrong,
these students could relieve the pressure that other students may have felt to be brilliant all
the time.
Detrimental peer behavior
To a large extent, the behaviors described as destructive were the opposite of those presented
above as good peer behaviors. However, it is important to distinguish between students who merely
fail to engage in beneficial behaviors and students who engage in behaviors that instructors reported
as having actively negative influences on peers.
17
• Dismissive or judgmental behavior was a common characteristic of a destructive peer.
Highly perceptive comments by high-ability students can be positive or negative depending
on the tone with which they were delivered. Students who perceive themselves to be better
than their peers sometimes discourage discussion in a number of ways: by talking down to
the rest of the class, by attempting to “score points” by demonstrating their intelligence, by
making comments with a “tone of authority,” or by attempting to engage in a dialog with the
instructor to the exclusion of peers.
• Rigidity of views may present difficulties. Good discussion involves being willing to admit
that you might be wrong, which is difficult for some students, especially on sensitive topics.
For example, some instructors reported that students with strong religious views had
particular difficulties discussing religious texts dispassionately.
• Rejection of peer learning was reported by some instructors, with the prevalence varying
widely across colleges. Such students wanted professors to tell them the “right” answers
rather than engaging in a group conversation. Several instructors described a student attitude
that might be represented as “I’m not paying all this money to listen to a bunch of 18-year-
olds.”
• Excessively loquacious peers (sometimes unintentionally) could stifle participation by other
students. Some such students always have to be the first to speak, others keep talking
endlessly (often about issues only tangentially related to the text under discussion), and some
repeat the same point over and over. One instructor referred to such a person as “annoying
boy.” (This behavior was more commonly reported among male students than females.)
Most instructors reported that they attempted to manage such situations with a private
conversation with the student. One mentioned a correlation of such disruptive behavior with
specific learning disabilities.
• Active or ostentatious disengagement that “exudes boredom” was reported by some
instructors. Examples included ignoring the class discussion, looking out the window, hostile
body language, checking text messages or email (or surfing the Web), and in one case
performing conspicuous acts of personal grooming during class. These actions were
distracting to other students and could foster or contribute to a more widespread negativity
toward the class.
While most instructors were able to relate examples of detrimental behavior, few reported
situations where this behavior had “destroyed” the class’s effectiveness. In some cases, the instructor
18
reported that he or she had succeeded in managing the behavior; in others the other students in the
class had marginalized the offending student. When, however, the detrimental behavior was
attributed to severe social awkwardness, students often shrugged it off fairly readily.
The quiet peer and the “nice” peer
A special category of peer is the student who consistently avoids contributing to class
discussions but who does not display active disengagement. Nearly every instructor had taught
many quiet peers and talked about them at some length. Some students are shy by nature and find it
very difficult to speak in front of others. Others have been taught to be passive in the classroom
setting. One common example was international students whose educational and cultural
backgrounds have conditioned them not to speak in class.
All instructors agreed that there were always a few quiet students in a class section. Most
believed that this was not usually a problem. However, there seemed to be a tipping point at which
too many quiet peers made it difficult or impossible to sustain a productive discussion. Many veteran
instructors had experienced at least one such section in their years of teaching.
Most instructors hesitated to find fault with quiet peers and seemed generally sympathetic.
Many described positive behavior by quiet peers: active listening, taking careful notes, and showing
clear, if silent, signs of engagement. Some noted that the quiet student’s occasional contributions
were often given disproportionate attention and impact by his or her peers: “when the wallflower
finally speaks up, it is more appreciated.” One instructor felt that the quiet but engaged students
often learned the most from their peers, carefully writing down everything that was said in class and
using it in their subsequent papers.
Another special category of peer that was described by some instructors is the student who is
“too nice” to disagree with a statement made by another student. The prevalence of this behavior
seems to vary dramatically across the three colleges. Although most instructors agreed that civility
was a positive and important norm of classroom demeanor, some students seemed to carry this too
far. “They don’t want to disturb the waters of the friendly community (so) they … let points of
intellectual disagreement pass or even avoid critically thinking about what someone is saying.”
Successful and unsuccessful class sections
A discussion class is more than a simple aggregation of its individual students. While a class
filled with students exhibiting good peer behavior (by the criteria above) would likely be a very
effective one and a class of destructive peers would almost surely be disastrous, all instructors agreed
that the impact of any individual peer or behavior depends crucially on the peer group within which
19
it occurs. A generally enthusiastic and active group can probably ignore or ostracize an actively
disengaged classmate who might disrupt a quieter or more marginally engaged class.
Moreover, there may be various “roles” played in a classroom discussion: initiator, lightning
rod, advocate, skeptic, joker, curmudgeon, peace-maker, etc. Just as 80 brilliant oboe players do not
make an orchestra, a class full of lightning rods, skeptics, or jokers is sure to fail. A successful core-
class section may require not just “good” individual peers, but also a mix of students who can fill the
necessary roles. One instructor pointed out that “We can’t have all questioners; we need to have
some answerers as well.” Another noted that students seem to settle into the roles of “first to speak,”
“first to respond,” “get involved later,” and “not speak at all unless called on” roles.
A mix of student interests (or prospective majors) seems desirable in a core class as well.
With a diversity of backgrounds and interests, each student may get a chance to be excited by the
material and shine in class on a different day. “X might be great at the philosophical text and less
good with the poetry, and that might be Y’s forte.” A diverse set of background experiences was also
thought desirable. One instructor said that in an ideal section “you have some kids from small
towns, some kids from cities. You have some kids who went to religious schools or are themselves
very religious and you have some who like to think of themselves on the cutting edge of atheism.”
The instructor also contributes heavily to the classroom mix of personalities. One role of the
instructor in a discussion is to step into an unfilled role: initiating conversation if others do not,
making a joke when levity is needed, doubting the consensus if skeptical viewpoints have gone
unspoken, or calming conflicts between forceful students.
Instructors, like their students, adopt some roles more naturally than others, so a naturally
skeptical instructor would complement a class lacking students to fill this role. Instructors differ in
their ability to manage various potential disfunctionalities in the classroom as well. Some instructors
thrive on lively, even heated, exchange while others might lose control of a class where discussion
became too spirited. Some instructors are more able to draw quiet students into the discussion than
others, for whom a quiet group of students might never get started.
Because of the complexities of classroom interaction, we tried to probe beyond individual
actions to learn about the group characteristics of successful and unsuccessful class sections.
Instructors were generally able to recall specific class sections that they viewed as more (and less)
successful than others. When asked what made sections successful, they often described the
contributions of individuals in the class, referring to some of the positive individual behaviors
described above. Many discussed the negative impacts of individuals’ dysfunctional behaviors as
leading to unsuccessful sections.
20
Analyzing the interviews, the most common definition of a good section seems to be one
that can sustain academically productive class discussion involving the majority of students. Toward
this end, the most important “role” that instructors commonly invoked for making sections work
was the student whose own contributions were of a kind that encouraged others to get excited about
the class and join the conversation. A few such positive peers could make a class successful. Those
whose class contributions (either through participation or active non-participation) trivialized the
class or discouraged peers from speaking were at the center of most unsuccessful sections.
In addition, some instructors commented on the importance of class bonding. It was not
necessary for students to be close personal friends, but feelings of familiarity and acceptance led to
“academic friendship” among members of successful classes.
Students were also more satisfied, comfortable, and confident if they felt that theirs was a
“good class.” This class self-image can be formed early in the year and may be difficult to reverse if it
becomes negative.
A concentric-ring model of peers
An image evolved during the course of our conversations that describes the initial
composition of a class section in terms of three concentric rings. In the center is an active and
engaged group—those naturally “good peers” who demonstrably take the enterprise seriously,
model good academic performance, and encourage their peers to join in the classroom intellectual
discourse. In the outer ring are any students who are predisposed to be actively disengaged or hostile
on the periphery. Lying between them in the middle ring is a potentially large group of initially
“uncommitted” students who are not disposed to strongly positive or strongly negative peer
behavior.
We discussed above many different characteristics of good and detrimental peers. These
underlying student characteristics vary on a multivariate continuum that resists facile reduction to a
single measure of “good-peer-ness.” Thus our attempt to depict this complex web of factors in a
single dimension must be viewed as a simple abstraction. Moreover, while there was consensus that
students vary considerably in their predisposition to be good peers, drawing any kind of borders
between good and neutral or neutral and detrimental is clearly arbitrary.
Nonetheless, the concentric-ring model provides a useful framework for describing two
closely related phenomena that we heard described by many instructors. The first is the “critical-
mass hypothesis” that a class is successful if the central group is large enough to sustain good class
discussion with occasional contributions from the middle group and despite potential distractions
from the periphery. This hypothesis is consistent with a static view of student behavior that is
21
determined mostly by personality and background and that does not change much within the core
class. If enough of the students exhibit good peer behavior, then the class will be successful for most
students even if those outside the center never become strong participants themselves.
The second is the “gravitational-attraction hypothesis” that students in the middle ring tend
to be drawn over time either into the center or toward the periphery. Successful classes are those in
which the center is large enough (and the periphery small enough) to attract a large share of the class
into positive peer engagement. This hypothesis emphasizes the dynamic evolution of student
behavior over the duration of the core class in response to the behavior of peers. Whereas the central
group sustains the class by itself under the critical-mass model, it does so in the gravitational model
by inducing additional students to become active participants.
Critical-mass hypothesis
All interviewees agreed that class discussions require a minimum critical mass of active
participants to be successful. The number of active contributors that instructors felt was necessary in
a class of 15–20 ranged from a low of three to a high of six or more. Those who viewed student
participation as being largely determined exogenously by their innate personalities tended to view
the success of a class in terms of having enough outgoing students to sustain lively interaction.
Having a critical mass of participants of varying levels of sophistication can serve to anchor
the conversation at a level that is accessible to all students (and so may encourage all to participate).
The mass may, for example, often “translate” for the benefit of weaker members of the class the
input of very strong students, whose contributions might otherwise not elicit responses (and who
might therefore cease to contribute). Equally, they may translate the input of weaker students by
adding evidence or refining terms, so that the whole class can benefit.
In terms of the concentric-ring model, students in the center group are consistent
contributors who can be counted on to be active in class. The peripheral group does not participate
positively. We characterized as the middle group those who contributed sporadically, either offering
an occasional comment regularly or varying between active and inactive participation depending on
the topic, their daily degree of preparation, or other factors.
According to the critical-mass hypothesis, a successful class is one with a sufficient number
of students whose engagement and personalities make them active, central participants. Detrimental
peer behavior was not frequently discussed by interviewees promulgating the idea of critical mass, so
the distinction between the districting peripheral group and the middle was less important.
22
Gravitational-attraction hypothesis
For many instructors, the success of the class seemed to depend on relative size of the center
and the outer ring and on the direction toward which this middle ring of students is pulled. Many
interviewees reminded us that the students entering these courses are 18 years old and are
experiencing college for the first time; they are “like unopened blossoms” in the phrase of one
instructor. They do not know the norms of behavior that are expected in a college classroom and
have not yet formed their academic identities. This means that their behavior is likely to evolve
considerably over their first year as these academic identities develop.
Peer pressure and instructor influences within the core classroom will play an important role
in teaching new college students behavioral norms and in helping them to define their identities as
students. This is one reason that we might expect strong peer effects to emerge from these classes. As
one instructor put it, by the end of the first month the students will have decided “whether it’s going
to be cool to play” the academic game.
In terms of the concentric-ring model, if most of the middle ring “buys in” and is attracted
toward the center then the majority of the class participates in discussions and the class is likely to be
successful. If the middle ring “opts out” toward the periphery then classroom interaction will be
strained because the number of active participants is small.
Many factors influence the middle ring of students. Having a sufficient number of actively
engaged participants can instill a norm of intellectually active behavior in these wavering, middle-
ring students, pulling them into the active group and making a highly successful class. When there
are too few students who are initially predisposed toward active engagement, it is harder to
demonstrate intellectually productive peer-with-peer interaction, and thus more difficult to establish
active discussion as a valuable norm. Moreover, if active peers are too sparse, they may be seen by
the majority of students as exemplary in a way that does not encourage imitation—they may be
viewed as simply smarter or better readers—and instead shuts down conversation.
With a solid and sizable center establishing a centripetal norm of academic engagement, it
may be possible for a class to shrug off the centrifugal influence of actively disinterested or even
hostile members. However, if a large proportion of the class is disengaged and non-participatory
then silence, indifference, or rejection of the class can become accepted as the norm, leaving the
small central group as the lone active participants in a less-than-successful class. Some instructors
reported that one sufficiently disruptive student had sometimes been enough to derail an entire class.
23
External affinities of middle-ring students with center or periphery students may also affect
whether they become active participants. One instructor noted a class with several athletes in which
excellent peers drew their teammates into positive roles.
Conclusions from faculty interviews
Despite the diversity of responses that we heard from our interviewees, we were able to draw
tentative conclusions about some aspects of peer effects in the core courses at our three institutions.
• Peers matter. Although many had not explicitly considered peer effects prior to our study,
all the instructors we interviewed believed that peers affect students’ learning. They were
particularly aware of peer effects through participation in classroom discussions, which
affects fellow students’ learning both of content and of the process of academic discourse.
• There are specific characteristics that make good peers. There was near-universal
consensus about the individual characteristics and behaviors through which good peers
enhance students learning.
o Personality and attitude are crucial. The most important peer characteristics
seemed to be one associated with students’ personalities and their attitudes toward
learning and toward the class. The most stellar peers were often A– or B+ students,
but they always behaved in ways that stimulated others’ interest in the class and
invited their active participation in discussions.
o Good academic quality is also important. The best peers were never C students and
often had A or A+ ability.
• A critical mass of strong peers is important. Many instructors highlighted the importance
of having a critical mass of positive peers who were capable of elevating the quality of
discourse to a high level. That is, the composition of the class as a whole matters as much as
the individual characteristics of members.
• Classmate role models are very important. Many students do not come into core class with
established academic identity. They establish it partially in the core class, and partially on
the basis of role models in that class. A central group of good peers can convince others that
it’s “cool to play” the academic game, drawing them into the center and promoting a
successful class.
• There are diverse roles to be played in a successful classroom, so a diversity of personalities
and backgrounds is good. Even though most instructors could describe an “ideal peer,” it
24
was not always clear that one needed, or even wanted, a whole class full of them. A class
was more than a simple sum of the individuals in it. While in an “ideal class,” all students
would be careful readers, excited about the course, and civil in demeanor, such an ideal class
would also need diversity: both questioners and answerers, both skeptics and consensus-
seekers, both quick responders and careful, deliberate thinkers.
• Males and females tend to play different roles in the classroom. We heard many
instructors characterize distinct academic personalities that they associated with male and
female students. Males were more aggressive and tended to focus on content; females were
more cooperative and tended to be more concerned with process. Excellent peers who were
identified by instructors were about equally divided between males and females. Those
identified as destructive peers were more often male.
Conclusions
Combining the evidence from our statistical analysis and the instructor interviews, we have
strong evidence that within a selective college, having a class of students with strong admission
credentials does not automatically lead to better learning by an individual student in the class, given
his or her own characteristics. This, in itself, is a strong conclusion. It casts doubt on the pedagogical
benefit of targeting merit-aid money toward students solely on the basis of academic credentials.
Of course, our study is limited by the fact that nearly all of the students we studied have
relatively high credentials. Thus, our conclusion may hold only within the usual range of applicants
and might not apply if a college were able to use merit aid to significantly raise its entire student
body. Nonetheless, it does suggest that a class drawn disproportionately from the upper tiers of a
college’s normal application pool is not likely to lead to greater learning for any individual student in
the class than one drawn evenly from the distribution as a whole.
The interviews lead us to believe that the reason for our weak statistical results is that we are
measuring the wrong peer characteristics. If peer attitudes and personalities have more important
influences on classmate learning, then we should be more successful in finding statistically
significant effects if we can measure those characteristics. College databases have only fragmentary
information that might help us examine the effects of peer attitudes and personalities.
If early-decision (ED) applicants are “better fits” for the college and have more positive
attitudes, then it is possible that having more classmates who applied ED would be beneficial. We
tested this hypothesis and found no positive evidence that more ED classmates affected our outcome
measures.
25
For two of our schools, we had a general admission-office rating, giving an overall
assessment to the applicant’s file. This rating is based partially on SAT scores and high-school
record, which are part of our database, and partially on letters of recommendations and student
essays, which we cannot directly quantify. By taking a “residual” from a regression of the admission-
office rating on our other measured applicant characteristics, we were able to construct an indirect
measure of the quality of the letters and essays, which might be a better gauge of student attitude and
fit for the college. Again, we found no statistically significant effect of having classmates with higher
admission-rating residuals.
In order to test for attitude-related peer effects statistically, it will probably be necessary to
have direct data on incoming students’ personalities and/or attitudes about college. This could be
done by surveys of entering students, either using instruments specially designed for this purpose or
perhaps by extracting relevant information from the freshman survey designed by the Cooperative
Institutional Research Program (CIRP) and administered by many colleges.
A small pilot study of three sections of Humanities 110 at Reed College was conducted in
2009–10. Although the sample was too small for statistical analysis, the results from before-and-after
surveys of students and instructors strongly corroborated the conclusions we drew from our
interview analysis. When asked to identify the peer behaviors that aided their learning, students
overwhelmingly selected the same attitude-based behaviors that our interviewed instructors
suggested were most important and consistently viewed academic prowess as less important.
The end-of-year survey asked instructors to identify which students in the class had engaged
in various peer behaviors. Those students who had engaged in several of the behaviors that students
felt were beneficial to learning were them identified as “beneficial peers.” Looking at the pre-course
attitudes reported by the beneficial peers, there was a strong tendency for the beneficial peers to have
had (1) a positive attitude toward the humanities course and (2) a reported pre-course expectation
that they would participate actively in class discussions.
The results of this small survey study suggest that a larger study that might measure students’
pre-college attitudes and expected class participation could shed additional light on the nature of
peer effects. In particular, such a study over multiple years could allow the examination of
subsequent student outcomes in relation to the distribution of peer attitudes and the reported quality
of the core-course peer interaction.
26
Tables
Table 1
Lewis & Clark Reed Whitman Title of course Inventing America Classical Humanities Antiquity and Modernity Duration and syllabus Two semester courses:
Fall is common syllabus; students choose different topic-based sections in spring
Year-long course; few students change sections
Year-long course; very few students change sections
Student credit 4 semester hours each semester
6 semester hours (1.5 Reed units) for each semester
4 semester hours each semester
Contact hours per week 3 hours of discussion; occasional optional lectures
3 hours of discussion; 3 hours of required lecture
3 hours of discussion; occasional optional lectures
Content Fall: Classical to modern, emphasizing political philosophy relating to founding of U. S. and to groups left out of founding process
Classical philosophy, history, and literature; Greece in fall and Rome in spring
Philosophy, history, and literature, focusing on classical in the fall and early modern/modern in the spring
Section size Discussion sections of 18−19 (slightly larger in early years).
Discussion sections of 14−17.
Discussion sections of 16−18.
Scheduling of sections Most at common afternoon time. A few at morning time for in-season athletes and students with afternoon labs.
Sections are offered at many times and days.
All sections meet at common morning time.
Section choice Students rank-order four preferred sections/instructors.
Students register for section/time without knowing instructor.
Students are randomly assigned to sections by registrar.
Criteria for evaluation Writing, class participation, and exams.
Writing, class participation, and exams. Single grade for full year.
Writing, oral reports, discussion participation, exams. Semesters graded separately.
27
Table 2
Dependent variable is GPA (excluding core) in:
Coefficient (standard error) on peer-mean predicted GPA Lewis & Clark Reed Whitman
SAT No SAT Full sample Full sample
All courses –0.183 –0.144 0.119 –0.0445 (0.119) (0.182) (0.151) (0.110)
First-year courses –0.215 0.134 –0.0580 –0.0143 (0.148) (0.231) (0.174) (0.129)
First- & second-year courses –0.201 0.129 –0.0840 –0.0399 (0.127) (0.196) (0.161) (0.113)
Narrow core-related courses –0.142 0.221 –0.0043 –0.0582 (0.161) (0.281) (0.174) (0.125)
Broad core-related courses –0.165 0.113 0.0824 –0.0896 (0.122) (0.182) (0.161) (0.112)
28
References
Ammermueller, Andreas, and Jörn-Steffen Pischke. 2006. Peer Effects in European Primary Schools: Evidence from PIRLS. In Working Paper. Cambridge, Mass.: National Bureau of Economic Research.
Angrist, Joshua D., and Kevin Lang. 2004. How Important are Classroom Peer Effects? Evidence from Boston's Metco Program. In Working Paper. Cambridge, Mass.: National Bureau of Economic Research.
Boozer, Michael A., and Stephen E. Cacciola. 2001. Inside the 'Black Box' of Project Star: Estimation of Peer Effects Using Experimental Data. In Discussion Paper. New Haven, Conn.: Yale University Economic Growth Center.
Burke, Mary A., and Tim R. Sass. 2006. Classroom Peer Effects and Student Achievement. In Working Paper. Tallahassee, Fla.: Florida State University Department of Economics.
Dills, Angela K. 2005. Does Cream-Skimming Curdle the Milk? A Study of Peer Effects. Economics of Education Review 24 (1):19-28.
Ding, Weili, and Steven F. Lehrer. 2006. Do Peers Affect Student Achievement in China's Secondary Schools? In Working Paper. Cambridge, Mass.: National Bureau of Economic Research.
Epple, Dennis, and Richard E. Romano. 1998. Competition between Private and Public Schools, Vouchers, and Peer-Group Effects. American Economic Review 88 (1):33-62.
Foster, Gigi. 2006. It's Not Your Peers, and It's Not Your Friends: Some Progress toward Understanding the Educational Peer Effect Mechanism. Journal of Public Economics 90 (8-9):1455-75.
Gaviria, Alejandro, and Steven Raphael. 2001. School-Based Peer Effects and Juvenile Behavior. Review of Economics and Statistics 83 (2):257-68.
Hanushek, Eric A., John F. Kain, Jacob M. Markman, and Steven G. Rivkin. 2003. Does Peer Ability Affect Student Achievement? Journal of Applied Econometrics 18 (5):527-544.
Henderson, Vernon, Peter Mieszkowski, and Yvon Sauvegeau. 1976. Peer Group Effects and Educational Production Functions. Ottawa: Economic Council of Canada.
Hoxby, Caroline. 2000. Peer Effects in the Classroom: Learning from Gender and Race Variation. In Working Paper. Cambridge, Mass.: National Bureau of Economic Research.
Kang, Changhui. 2007. Classroom Peer Effects and Academic Achievement: Quasi-Randomization Evidence from South Korea. Journal of Urban Economics 61 (3):458-495.
Kremer, Michael, and Dan Levy. 2008. Peer Effects and Alcohol Use among College Students. Journal of Economic Perspectives 22 (3):189-206.
Lefgren, Lars. 2004. Educational Peer Effects and the Chicago Public Schools. Journal of Urban Economics 56 (2):169-191.
Lyle, David S. 2007. Estimating and Interpreting Peer and Role Model Effects from Randomly Assigned Social Groups at West Point. Review of Economics and Statistics 89 (2):289-299.
Manski, Charles F. 1993. Identification of Endogenous Social Effects: The Reflection Problem. Review of Economic Studies 60 (3):531-542.
McEwan, Patrick J. 2003. Peer Effects on Student Achievement: Evidence from Chile. Economics of Education Review 22 (2):131-41.
Parker, Jeffrey, James Grant, Jan Crouter, and Jon Rivenburg. 2010. Classmate Peer Effects: Evidence from Core Courses at Three Colleges. Portland, Ore.: Reed College.
Rangvid, Beatrice Schindler. 2003. Educational Peer Effects: Regression Evidence from Denmark with PISA2000 Data. Copenhagen: AKF, Institute for Local Government Studies.
29
Robertson, Donald, and James Symons. 2003. Do Peer Groups Matter? Peer Group versus Schooling Effects on Academic Attainment. Economica 70 (277):31-53.
Rothschild, Michael, and Lawrence J. White. 1995. The Analytics of the Pricing of Higher Education and Other Services in Which the Customers are Inputs. Journal of Political Economy 103 (3):573-586.
Royston, Patrick. 2005. Multiple Imputation of Missing Values: Update. Stata Journal 5:188-201. Rubin, Donald B. 1996. Multiple Imputation after 18+ Years. Journal of the American Statistical
Association 91 (434):473-489. Sacerdote, Bruce. 2001. Peer Effects with Random Assignment: Results for Dartmouth Roommates.
Quarterly Journal of Economics 116 (2):681-704. Stinebrickner, Ralph, and Todd R. Stinebrickner. 2006. What Can Be Learned about Peer Effects
Using College Roommates? Evidence from New Survey Data and Students from Disadvantaged Backgrounds. Journal of Public Economics 90 (8-9):1435-54.
———. 2008. The Causal Effect of Studying on Academic Performance. B.E. Journal of Economic Analysis & Policy 8 (1: Frontiers):Article 14.
Vandenberghe, V. 2002. Evaluating the Magnitude and the Stakes of Peer Effects Analysing Science and Math Achievement across OECD. Applied Economics 34 (10):1283-90.
Winston, Gordon C., and David J. Zimmerman. 2004. Peer Effects in Higher Education. In College Choices: The Economics of Where to Go, When to Go, and How to Pay for It, edited by C. Hoxby. Chicago: National Bureau of Economic Research and University of Chicago Press.
Zimmer, Ron. 2003. A New Twist in the Educational Tracking Debate. Economics of Education Review 22 (3):307-315.
Zimmer, Ron W., and Eugenia F. Toma. 2000. Peer Effects in Private and Public Schools across Countries. Journal of Policy Analysis and Management 19 (1):75-92.
Zimmerman, David J. 2003. Peer Effects in Academic Outcomes: Evidence from a Natural Experiment. Review of Economics and Statistics 85 (1):9-23.